Analytical Validation of a Deep Neural Network Algorithm for the Detection of Ovarian Cancer.

PURPOSE: Early detection of ovarian cancer, the deadliest gynecologic cancer, is crucial for reducing mortality. Current noninvasive risk assessment measures include protein biomarkers in combination with other clinical factors, which vary in their accuracy. Machine learning can be applied to optimizing the combination of these features, leading to more accurate assessment of malignancy. However, the low prevalence of the disease can make rigorous validation of these tests challenging and can result in unbalanced performance. METHODS: MIA3G is a deep feedforward neural network for ovarian cancer risk assessment, using seven protein biomarkers along with age and menopausal status as input features. The algorithm was developed on a heterogenous data set of 1,067 serum specimens from women with adnexal masses (prevalence = 31.8%). It was subsequently validated on a cohort almost twice that size (N = 2,000). RESULTS: In the analytical validation data set (prevalence = 4.9%), MIA3G demonstrated a sensitivity of 89.8% and a specificity of 84.02%. The positive predictive value was 22.45%, and the negative predictive value was 99.38%. When stratified by cancer type and stage, MIA3G achieved sensitivities of 94.94% for epithelial ovarian cancer, 76.92% for early-stage cancer, and 98.04% for late-stage cancer. CONCLUSION: The balanced performance of MIA3G leads to a high sensitivity and high specificity, a combination that may be clinically useful for providers in evaluating the appropriate management strategy for their patients. Limitations of this work include the largely retrospective nature of the data set and the unequal, albeit random, assignment of histologic subtypes between the training and validation data sets. Future directions may include the addition of new biomarkers or other modalities to strengthen the performance of the algorithm.

Using the Integrated Genome Viewer to reveal amplicon-derived polymorphism enriched at the phenylthiocarbamide locus in the teaching lab.

Due to its distinct phenotype and relatively simple inheritance pattern, the phenylthiocarbamide (PTC) loci is frequently utilized in teaching laboratories to demonstrate genetic concepts such as Mendelian inheritance and population genetics. We have developed a next-generation sequencing and bioinformatics approach to analyze the PTC gene locus to reveal single nucleotide polymorphism (SNP) variation at nucleotide position 785 that predicts tasting ability in humans. Here students purify DNA from their own cheek cells, perform polymerase chain reaction (PCR) amplification of the PTC gene followed by cleaved amplified polymorphic sequence (CAPS) testing. Students perform a second PCR on the PTC loci using high-fidelity Taq to create bar-coded amplicons for next-generation sequencing on the Ion Torrent Personal Genome Machine. Bioinformatic verification reveals polymorphic variation by aligning the entire class PTC PCR fragment sequence to the human gene using Bowtie2 and visualizing the results in the Integrated Genome Viewer. This exercise presents a learning opportunity for students to use next-generation sequencing to predict their own PTC taste sensitivity phenotype coupled with the standard CAPS method. This approach brings the PTC teaching method into the genomics era.

Adaptive Radiation Genomics of Two Ecologically Divergent Hawai'ian Honeycreepers: The 'akiapola'au and the Hawai'i 'amakihi.

The Hawai'ian honeycreepers (drepanids) are a classic example of adaptive radiation: they adapted to a variety of novel dietary niches, evolving a wide range of bill morphologies. Here we investigated genomic diversity, demographic history, and genes involved in bill morphology phenotypes in 2 honeycreepers: the 'akiapola'au (Hemignathus wilsoni) and the Hawai'i 'amakihi (Chlorodrepanis virens). The 'akiapola'au is an endangered island endemic, filling the "woodpecker" niche by using a unique bill morphology, while the Hawai'i 'amakihi is a dietary generalist common on the islands of Hawai'i and Maui. We de novo sequenced the 'akiapola'au genome and compared it to the previously sequenced 'amakihi genome. The 'akiapola'au is far less heterozygous and has a smaller effective population size than the 'amakihi, which matches expectations due to its smaller census population and restricted ecological niche. Our investigation revealed genomic islands of divergence, which may be involved in the honeycreeper radiation. Within these islands of divergence, we identified candidate genes (including DLK1, FOXB1, KIF6, MAML3, PHF20, RBP1, and TIMM17A) that may play a role in honeycreeper adaptations. The gene DLK1, previously shown to influence Darwin's finch bill size, may be related to honeycreeper bill morphology evolution, while the functions of the other candidates remain unknown.

Whole-genome bisulfite sequencing with improved accuracy and cost.

DNA methylation patterns in the genome both reflect and help to mediate transcriptional regulatory processes. The digital nature of DNA methylation, present or absent on each allele, makes this assay capable of quantifying events in subpopulations of cells, whereas genome-wide chromatin studies lack the same quantitative capacity. Testing DNA methylation throughout the genome is possible using whole-genome bisulfite sequencing (WGBS), but the high costs associated with the assay have made it impractical for studies involving more than limited numbers of samples. We have optimized a new transposase-based library preparation assay for the Illumina HiSeq X platform suitable for limited amounts of DNA and providing a major cost reduction for WGBS. By incorporating methylated cytosines during fragment end repair, we reveal an end-repair artifact affecting 1%-2% of reads that we can remove analytically. We show that the use of a high (G + C) content spike-in performs better than PhiX in terms of bisulfite sequencing quality. As expected, the loci with transposase-accessible chromatin are DNA hypomethylated and enriched in flanking regions by post-translational modifications of histones usually associated with positive effects on gene expression. Using these transposase-accessible loci to represent the cis-regulatory loci in the genome, we compared the representation of these loci between WGBS and other genome-wide DNA methylation assays, showing WGBS to outperform substantially all of the alternatives. We conclude that it is now technologically and financially feasible to perform WGBS in larger numbers of samples with greater accuracy than previously possible.